multi-operator fairness in transparent ran sharingspeed.cis.nctu.edu.tw/~ydlin/motf.pdf · 2018. 9....
TRANSCRIPT
-
1
Multi-operator Fairness in Transparent RAN Sharing¹Ying-Dar Lin, ¹
,²Hsu-Tung Chien, ²Hsien-Wen Chang and ²Chia-Lin Lai
¹Dept. of Computer Science, National Chiao Tung University, Hsinchu, Taiwan
²Information and Communications Research Laboratories, Industrial Technology Research Institute, Hsinchu, Taiwan
[email protected], [email protected], [email protected], [email protected]
Abstract—Radio access network (RAN) sharing attracts much
attention from telecom operators. However, current
mechanisms applied to RAN sharing do not consider fairness
among operators such that the RAN is not fairly shared, and
either under- or over-utilized. Fairness cannot be guaranteed
among operators because resources are distributed on a first-
come-first-serve basis. “Soft-partition with Blocking and
Dropping” (SBD)” is proposed to offer inter-operator fairness
based on a “soft-partition” concept. Subscribers to an operator
may overuse the predefined service-level-agreement (SLA)
when the shared RAN is under-utilized, but become blocked or
even dropped when over-utilized. According to our simulation
results, SBD maintains inter-operator fairness at 0.997, even
better than hard-partition at 0.98 and much better than no-
partition at 0.6. Meanwhile, it retains high utilization of
shared RAN at 98%. SBD reduces blocking rate from 35% of
hard partition to almost 0%, while controlling dropping rate at
5%.
Keywords—radio access networks; RAN sharing; soft
partition; fairness; utilization; blocking rate; dropping rate
I. INTRODUCTION
RAN sharing is an inevitable trend for telecom
operators for addressing upcoming data surges with
minimum investment in CAPEX and OPEX [1][2][3].
Transparent RAN sharing, which means RAN sharing
achieved through an intermediate mechanism with least
configuration to BSs and core networks, is especially
attractive to operators for the following reasons: 1). It is
easier and more cost effective for operators to set up
RAN sharing with existing infrastructure; 2). RAN
sharing can be managed by a third-party operator to
assure the independence and fairness among sharing
operators [4][5]. To achieve transparent RAN sharing,
RAN Proxy (RANP) was proposed in our earlier work [6]
with comprehensive emulation results.
To make RAN sharing more practicable, the fairness
and efficiency in allocating resource among sharing
operators need further consideration. RAN resource is
limited, especially the capacity of a base station (BS) in
terms of number of served user equipments (UEs) and
available bandwidth. For example, in LTE the RCC-
connected [7] users for a macro cell is more than 1000,
and the number may decrease to 50-100 for a typical 5G
small cell [8][9] considering its smaller coverage. As for
bandwidth, it may be limited by the bottleneck in the
entire backhaul network. Although LTE provides
sophisticated mechanisms for an operator to coordinate
UEs and their bandwidth within its own network, the
coordination among operators is still immature. On one
hand, if there is no coordination, and any new UE is
allowed to attach regardless of its operator until the
shared BS reaches its capacity limit, very large operators
with large subscriber bases may consume most
resources, and the UEs of smaller operators may be
of unacceptable quality of experience (QoE). On the
other hand, if the resource is reserved for each
operator according to predefined inter-operator
agreement, the QoE of UEs from each operator is
guaranteed, at the cost of reduced BS utilization if
any operator doesn’t fully use the resource reserved
for it, which causes reduced traffic as well as revenue
for the third party providing the RAN sharing service
[2]. How a trade-off between the conflicting
requirements becomes the key to the success of RAN
sharing.
In this paper, we propose a “Soft-partition with
Blocking and Dropping” (SBD) to control the
fairness among different operators based on “Soft-
partition” concept, which means that telecom
operators can use resource more than what is agreed
among operators whenever there is still available
resource. However, once the network is fully loaded,
the overusing operator will return the overused
resource according to a predefined agreement. In this
way, the utilization of RAN resources can be
increased while the requirements of each operator can
be satisfied. In addition, the blocking and dropping
rate for each operator can be maintained in at an
acceptable level.
The rest of this work is organized as follows.
Section II provides background and related works. In
Section III the problem is formulated, followed by
the detailed design of SBD mechanism in Section IV.
Numerical results of SBD mechanism, including
fairness among multiple operators, utilization of a
shared BS and blocked/dropping rate of resource
requests, are provided in Section V, followed by
conclusions and future work in Section VI.
II. INTER-OPERATOR CONTROL: RELATED
WORKS
Nokia [10] provides a solution for radio resource
management in a roaming scenario among multiple
operators in 3G networks. A Control-Plane-based
Hard Partition is used to fix the resource in CN.
Qualcomm [11] uses both Control-Plane and Data-
Plane in its solution. In Data-Plane, a mechanism
called “Erase Packet” detects which packet remains
in the queue for the longest time and deletes it to
release the radio resource when a BS runs out of
resource. After resource re-organization Control-
Plane can decide whether to accept a new request or
not.
2018 Workshop on Computing, Networking and Communications (CNC)
978-15386-3652-7/18/$31.00 ©2018 IEEE 259
-
2
The attributes of the above related works, including
location in the network, either based on Control or Data
Plane, on Soft or Hard partition, supporting UE and/or
bearer admission control, are listed in Table 1, together
with the proposed SBD mechanism. As can be seen in
Table 1, only SBD applies soft partition for BS resource
utilization.
III. PROBLEM FORMULATION
Inter-operator policy is the agreement among
operators for resource coordination, which is subject to
the constraint as
∑ �� �������_� � ∑ �� ��
�����_� � 1,
(1) where Pinter_ue
= {��
�����_�} and Pinter_bw
= {��
�����_�}
describes the percentage of UEs and bandwidth resource
allocated to i-th operator, respectively. Given BSlimit_ue
,
the maximum number of served UEs for a BS, the i-th
operator’s quota in number of served UEs is determined
by the inter-operator policy and calculated as pinter_ue
i *
BSlimit_ue
. Similarly, given BSlimit_bw
, the maximum
bandwidth of a BS, the i-th operator’s quota in bandwidth
usage is calculated as pinter_bw
i * BSlimit_bw
.
The Fairness in terms of number of served UEs and
bandwidth usage is defined as
�� � ∏ ������
�������_��
∗�������_��
, (2)
and
�� � ∏ �����
�������_�
∗�������_�
, (3)
respectively, where !�
� and !�
� are the quantities of in-
use UE and bandwidth of i-th operator. According to the
definition, the more �� (or �� ) approaches one, the more actual resource assignment is aligned to inter-
operator policy, and fairer the system is considered.
The utilization of a shared BS in terms of number of
served UEs and bandwidth usage is defined and
constrained as
"� �#
�������_��
∑ �� !�
� ∗ 100% & 1, (4)
and
"� �#
�������_�
∑ �� !�� ∗ 100% & 1, (5)
respectively, and can be used to evaluate the resource
utilization efficiency of different methods.
As suggested by the name SBD, portion of the new
resource requests will be blocked, while portion of the in-
use resource will be dropped, and QoE of UEs will
inevitably be affected. The impact can be evaluated by
the blocking rate of UE Attach and the bandwidth
request, which are defined as
'��()*+_� �
�()*+�,-./����0(1��0*23�4����
5)�0(3�4����6 100%, (6)
and
'��()*+_� �
�()*+�,�0�,�,�2
5)�0(�0�,�,�23�4����6 100%. (7)
The impact is also evaluated by dropping rate of in-use
UEs and bandwidth defined as
'�,�)�_�
�8�)�-.
5)�0(3�4����6 100%, (8)
and
'�,�)�_�
��0�,��,�23�(�0��
5)�0(�0�,�,�23�4����6 100%. (9)
IV. PROPOSED BLOCKING AND DROPPING
COORDINATOR
The proposed Soft-partition with Blocking and
Dropping (SBD) is integrated with RANP as shown
in Fig. 1. It provides a two-stage control on two types
of fundamental resource in the shared BS. The first
stage is UE admission control. After being triggered
by a UE Initial Attach request, SBD decides to accept
or block the request, or drop some in-service UE
whose operator has overused its quota, i.e.,
predefined number of UEs. The second stage is
bandwidth grant control. By listening to E-RAB
messages [8], SBD detects E-RAB Request and E-
RAB Modify messages and decides to accept or
block bearers, or drop some existing bearers whose
operator has overused its quota, i.e., predefined
guaranteed bandwidth. More details are described in
next two subsections.
Fig. 1. Soft-partition with Blocking and Dropping
Design
4.1 UE Admission control
Fig. 2 shows how UE admission control in SBD
works in detail. When a packet arrives, SBD inspects
and identifies its type. If it is an Initial Attach
request, SBD further identifies the UE’s operator, and
then retrieves the current status of BSfree_ue
(number of
additional UEs that BS can serve) and sue
i (number of
in-service UEs of i-th operator). With this
information, SBD decides to accept or block the
Initial Attach request by checking if sue
i is larger than
the i-th operator’s quota or not. If BSfree_ue
is greater
than zero, SBD accepts all the requests. However, if
BSfree_ue
equals zero, and sue
i is larger than the i-th
operator’s quota, the new request will be blocked. On
the other hand, if sue
i is smaller than the i-th
operator’s quota, the UE dropping procedure (Fig. 3)
starts so that other operators return the overused
resource, and then SBD can accept the request. In the
UE dropping procedure, a dropping list is maintained,
and one candidate is selected each time. First, SBD
looks for the operator which mostly overuses by
9:;�!�� < =�
�����_��∗ >?(�@��_�A. (10)
When the operator is identified, the idle UE will be
selected. If there is no idle UE, the procedure
continues to detect the status of UEs’ bearers to find
a UE without any extend dedicated bearer, i.e. with
only a default bearer. If there is more than one UE,
2018 Workshop on Computing, Networking and Communications (CNC)
260
-
3
Table 1. Comparison of related works
Papers / Patents Location Data / Control
Plane
Hard / Soft
Partition
UE Admission
Control
Bearer Admission
Control Method
Roaming Based [10] Core CP Hard No No - CN Control
Asymmetric RAN
Resource Allocation
[11]
Base Station CP/DP Hard No Yes
- Blocking
- Weighted Queuing
- Erase Packet
SBD RAN Proxy CP Soft Yes Yes - Blocking and Dropping
- Temporarily Over Quota
the one with longest connection time will be selected
and added to the dropping list. If every UE has
extended dedicated bearers, the one with least QCI
priority will be selected and added to the dropping list.
Fig. 2. UE Admission Control Procedure
Fig. 3. UEs Dropping Procedure
4.2 Bandwidth Grant Control
The detail of bandwidth grant control in SBD is
shown in Fig. 4. When a packet arrives, SBD inspects it
to identify the operator it belongs and to ensure it is an
E-RAB Request or an E-RAB Modify message. SBD
then retrieves the current status of BSfree_bw
(the
bandwidth available to allocate) and sbw
i (bandwidth
allocated to i-th operator), and decides to accept or
block the E-RAB message. If BSfree_ue
is greater than
zero, SBD accepts all the requests. However, if BSfree_ue
equals zero, and sbwi larger than the i-th operator’s
quota, the new request will be blocked. On the other
hand, if sbw
i is smaller than the i-th operator’s quota, the
bandwidth release procedure (Fig. 5) starts so that other
operators return the overused resource, and then SBD
can accept the request. In the Bandwidth Release
procedure, another dropping list is maintained. Because
of the varying size in different bandwidth requests, the
procedure iterates until sufficient resource is released.
First, SBD looks for the operator which mostly
overuses by
9:;�!�� < =�
�����_�� ∗ >?(�@��_�A. (11)
When the operator is identified, the UEs with extend
dedicated bearers will be detected. If every UE has
extend dedicated bearers, the one with least QCI
priority will be selected and added to the dropping list.
Otherwise, if there are UEs with the same QCI, the one
with the longest connection time will be selected for the
dropping list. If there are no UEs with dedicated
bearers, the procedure continues to detect the status of
UEs’ bearers to find the candidate with longest-time
connection. Finally, the procedure calculates the
released resource. If the released resource is not
enough, the procedure resumes until sufficient resource
has been returned to the corresponding operator.
Fig. 4. Bandwidth Grant Control Procedure
Fig. 5. Bandwidth Release Procedure
2018 Workshop on Computing, Networking and Communications (CNC)
261
-
4
V. NUMERICAL RESULTS
5.1 Simulation Setup
For simplicity, there are only two operators in the
simulations, and they agree to evenly share the resource
as
pinter_ue
i = p
inter_bwi = 0.5, where i=1, 2. (12)
The maximum number of served UEs and quantity of
allocated bandwidth for the shared BS is 100 UEs and
100 Mbps. According to (9), the quota for each operator
is 50 UEs and 50 Mbps. The distribution of arriving
UEs and bandwidth request from operator-i follow a
Poisson distribution with arrival rate B�
� and B�
� ,
respectively. Full loading is assumed in the fairness
evaluation experiment, while in utilization and
blocking/dropping rate evaluation experiments, the
system is not fully loaded. Every simulation condition
simulates duration of 500 minutes, during which in-use
resource is released (UE or bandwidth) whenever use
time reaches 250 minutes.
5.2 Results
The performance of SBD can be evaluated in terms
of 1) resource fairness among operators 2) total
resource utilization of the shared BS and 3) the
blocking and dropping rate of new resource requests
and in-use resources, respectively. The evaluation
results are described and discussed in the following
subsections.
5.2.1 Fairness among Multiple Operators
Fairness is defined as in (2) and (3). When there is
no control at all (i.e. no partition), UEs arriving first
become attached until the shared BS reaches the
maximum number of served UE, and so does
bandwidth request. For hard partition mechanism, each
operator can only use the resources allocated to it and
overusing a resource is strictly forbidden. With SBD,
on the other hand, operators are allowed to overuse a
resource, as described in Section IV. In Fig. 6, we
depict fairness in terms of (a) Number of UEs and (b)
bandwidth usage, using the three mechanisms (without
partition, hard partition, and SBD) with respect to
varying UE arriving ratio which is subject to the
constraint as ∑ ��C# B�
�
� 5, and ∑ ��C# B�
�
� 5E .
Intuitively, when there is no partition, the operator with
higher UE ratio takes up more resource, and the fairness
decreases. As for hard partition and SBD, fairness is
close to 1 (perfectly fair), and the small deviation is
because the shared BS is not fully loaded. SBD is
slightly better than hard partition because with SBD
operators can put more UEs into the system in some
occasions.
5.2.2 Utilization of Shared BS The Utilization of a shared BS is defined as in (4)
and (5). In Fig. ˙, we provide utilization of shared BS in
terms of (a) Number of UEs and (b) bandwidth usage,
using the two mechanisms (hard partition and SBD)
with respect to varying UE arriving ratio which is
subject to the constraint as ∑ ��C# B�
�
� 2, and
∑ ��C# B��
� 1E . In the experiment, B#
� and B#� are
fixed, and BG
�and BG
�are gradually decreased..
As shown in Fig. 7(a), the utilization of HP is closed to
SBD when the arrival rate is identical, but it is only
70% when HI��
HJ�� � 5. Furthermore, the utilization of SBD
is always around 95%. In Fig. 7(b), SBD improves from
63% to 98% in utilization of bandwidth usage when HI�
HJ� � 5 . The deviation of utilization from 100% is
because the system is not kept in full-loading status all
the time. In summary, SBD greatly improves the
utilization of shared BS. If RANP is operated by a
third-party operator, and the revenue is utilization-
based, SBD has stronger economic benefits.
5.2.3 Blocking Rate and Dropping Rate
The Blocking rate and Dropping rate is defined as in
(6)-(9). In Fig. 8, we provide blocking rate and
dropping rate of i-th operator in terms of (a) Number of
UEs and (b) bandwidth usage, using the two
mechanisms (hard partition and SBD) with respect to
varying UE arriving ratio subject to the constraints as
∑ ��C# B�
�
� 2, and ∑ ��C# B�
�
� 1E . In the
experiment, B#
�and B#
� are fixed, and BG
�and BG
� are
gradually decreased. This means the resources for
Number of UEs and bandwidth of operator 2 can be
under-utilized because of decreasing usage. Therefore,
as can be seen that in Fig. 8 (a), '#�()*+_�
is reduced
from 35% of HP to almost 0% of SBD because of the
UE dropping feature of SBD. For operator 2, 'G�()*+_�
under HP and SBD mechanisms have the same trend,
since the operator 2 gradually decreases its
requirements of Number of UEs. Thus, the available
resources can always satisfy operator 2’s needs. As for
the dropping rate, it can be seen that HP has no
dropping UEs for operators 1 and 2 (i.e., '#,�)�_�
and
'G,�)�_�
of HP), since it will not overuse the resource,
while SBD may have about a 5% dropping rate for
operator 1 (i.e., '#,�)�_�
of SBD), since operator 1
overused the resource borrowed from the operator 2,
and then returns the resources temporarily borrowed
when operator 2 needs it. The trend of
'#,�)�_�
resembles an inverted U-shaped curve because
when HI��
HJ�� � 1, the overused condition does not occur
frequently, but when HI��
HJ�� � 2 , operator 1 is usually
overused while operator 2 also has many requests,
operator 1 should return resources frequently. When the
arrival rate of operator 2 decreases, the resources
operator 1 should return also decrease. Similar
phenomena occur in Fig. 8 (b) for bandwidth allocation
between operator 1 and operator 2. In summary, SBD
can improve the blocking rate efficiently when the
shared BS is under-utilized, and control the dropping
rate at around 5%. On the other hand, when shared BS
is over-utilized and the arrival rates of operators are
2018 Workshop on Computing, Networking and Communications (CNC)
262
-
5
close, the hard partition is preferred to avoid
unnecessary dropping occurring.
VI. CONCLUSIONS AND FUTURE WORK
Soft-partition with Blocking and Dropping (SBD) is
proposed in transparent RAN sharing to control the
fairness among different operators based on a “Soft-
partition” concept. In SBD, telecom operators can
utilize resources more than what is agreed among the
operators when the shared BS is under-utilized.
However, SBD only allows overusing whenever there is
no impact on the other operators, i.e. SBD monitors the
usage of the resources for each involved operator and
dynamically adjusts the available resources among the
operators by blocking or even dropping. According to
our simulation results, SBD not only makes sure of
fairness between sharing operators but also keeps the
high utilization of shared BS around 100%.
Furthermore, SBD obviously reduces blocking rate
from 35% of hard partition to almost 0% when the
shared BS is under-utilized, and control the dropping
rate at around 5%.
In the future work, an enhanced algorithm and scale
down feature will be added to SBD to optimize the
blocking and dropping rates, and mathematical analysis
will be included to validate the extended SBD scheme.
A comprehensive model will be developed to analyze
the performance of SBD in terms of resource
utilization, blocking rate and dropping rate to prove the
validity of SBD scheme.
(a) Number of UEs
(b) Bandwidth Usage
Fig. 6. Fairness among Multiple Operators
(a) Number of UEs
(b) Bandwidth Usage
Fig. 7. Utilization of Shared Base Station Comparison
(a) Number of UEs
(b) Bandwidth Usage
Fig. 8. Blocking Rate and Dropping Rate
REFERENCE [1] T. Frisanco, P. Tafertshofer, P. Lurin and R. Ang,
“Infrastructure sharing and shared operations for
mobile network operators From a deployment and
operations view,” IEEE Network Operations and
Management Symposium, April 2008.
[2] GSMA Head Office, “Mobile Infrastructure
Sharing,” Accedian RAN Sharing White Paper,
Sep. 2012.
[3] Nokia Solutions and Networks, “Network
Sharing: Delivering mobile broadband more
efficiently and at lower cost,” Nokia Network
Sharing White Paper, 2014.
[4] Accedian Network, “RAN Sharing Solutions:
Network Performance Monitoring,” Accedian
RAN Sharing White Paper, Q2 2015.
[5] J. Markendahl and A. Ghanbari, "Shared
smallcell networks multi-operator or third party
solutions - or both?," Modeling & Optimization
in Mobile (WiOpt), May 2013.
[6] Y. D. Lin, H. S. Chien, H. W. Chang et al.,
“Transparent RAN Sharing of 5G Small and
Macro Cells”, IEEE Wireless Communication
Magazine, to appear. Available in
(https://goo.gl/pYmmid).
[7] Y. D. Lin, H. S. Chien, H. W. Chang et al.,
“Transparent RAN Sharing of 5G Small and
Macro Cells”, IEEE Wireless Communication
Magazine, to appear. Available in
(https://goo.gl/pYmmid).
[8] L. Frenzel, “Understanding The Small-Cell And
HetNet Movement,” ELECTRONIC DESIGN,
Sep. 2013.
[9] X. Ge, S. Tu, G. Mao et al., “5G Ultra-Dense
Cellular Networks,” IEEE Wireless
Communications, Volume: 23, Feb. 2016.
[10] K. Johansson, M. Kristensson and U. Schwarz,
“Radio resource management in roaming based
multi-operator WCDMA networks,” IEEE 59th
Vehicular Technology Conference, May 2004.
[11] A. Gogic and G. B. Horn, “Asymmetric Radio
Access Network (RAN) Resource Allocation in
RAN Sharing Arrangement,” US patent,
20140029529, Jan. 2014.
2018 Workshop on Computing, Networking and Communications (CNC)
263
2018-06-19T16:05:39-0400Certified PDF 2 Signature